The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The teamSearch function in SKILL.md accepts raw user input via the query parameter and interpolates it directly into the LLM message stream without specific delimiters or sanitization.
Ingestion points: User-supplied query parameter in the teamSearch function (SKILL.md).
Boundary markers: The provided code snippet does not implement boundary markers or instructions to ignore embedded commands.
Capability inventory: The skill manifest requests Read, Write, and Edit permissions, providing an attack surface if the model is manipulated via the search query.
Sanitization: No sanitization or input validation is present in the implementation logic.
[DATA_EXPOSURE_&_EXFILTRATION]: The skill uses process.env for API key management in its TypeScript implementation, which is a recommended practice for preventing accidental credential exposure.
[COMMAND_EXECUTION]: The skill includes a bash snippet for API key verification using curl. This operation targets the official api.perplexity.ai endpoint and is used for status validation rather than remote code execution.

perplexity-enterprise-rbac